One was an orchestral piece inspired by music from the Star Wars movies, but composed by an AI program from Belgian startup AIVA that — of course — relies on Nvidia chips. The music went over big with the crowd of AI geeks attending the Neural Information Processing Systems Conference, known as NIPS, including some giants in the field like Nicholas Pinto, head of deep learning at Apple, and Yann LeCun, director of AI Research at Facebook.

LeCun was quoted saying the Star Wars bit was “a nice surprise.”

Huang’s other surprise was a bit more practical, and showed just how competitive the AI chip market niche has become. Analysts say sales of specialized microprocessors for use with AI programs like machine learning and image recognition will grow astronomically from about $500m last year to $20bn to $30bn within five years. As the kinds of graphics chips that were first popular with video gamers have turned out to be among the best-suited for AI programming, too, Nvidia is in the lead, chased by Intel, Advanced Micro Devices, and a host of others. Intel is developing AI-focused chips from its 2016 acquisition of Nervana Systems, while AMD has an entirely new chip design tuned for AI apps called Vega on the market this year.

Hoping to stay at the front of the pack, Huang had a new graphics card called the Titan V to show off in Long Beach. Suitable for installing in ordinary PCs, the Titan V contains Nvidia’s latest “Volta” chip design with some 21 billion individual transistors. Priced at $3,000, the Titan V comes just eight months after Nvidia unveiled its Titan Xp design based on its earlier “Pascal” chips. The new card is a little more than double the price but offers up to nine times faster performance on major AI apps like Google’s TensorFlow software, Amazon’s MXNet and Facebook’s Caffe 2, Nvidia said. The older Nvidia card competed head on with AMD’s similarly priced Vega Frontier Edition, but the Titan V may prompt the need for a higher-priced, higher-performing AMD countermove if it catches on.

Read Full Article